The present invention relates generally to complementary metal oxide semiconductor (CMOS) static random access memories (SRAM), and more particularly to a single-ended read, differential write CMOS, SRAM cell including two inverters connected in a regenerative feedback circuit, wherein each inverter includes a pair of complementary field effect transistors (FETs), and FETs of the same type in each inverter have differing gate region widths and/or drive currents.
FIG. 1 is a block diagram of a prior art CMOS SRAM 10 including many memory cells. Each cell contains a pair of inverters connected in a regenerative feedback circuit. Each of the inverters includes two complementary FETs. The cells of SRAM 10 are addressed by word write/read logic network 12, bit write logic network 14, and bit read logic network 16. Typically, SRAM 10 is a matrix of cells including M-words, each having N-bits, where M is 0, 1 . . . i . . . M, and N is 0, 1 . . . k . . . N; M and N are typically powers of two minus the quantity one, e.g., 255 and 1,023, respectively. Word write/read logic network 12 includes M output terminals, word0, word1 . . . wordi . . . wordM, one of which is provided for each of the M words of SRAM 10. Bit write logic network 14 includes N complementary output terminals, nbit0, nbit1 . . . nbitk . . . nbitN and N non-complementary output bit terminals, bit0, bit1 . . . bitk . . . bitN. Bit read logic network 16 includes N complementary input terminals, nbit0, nbit1 . . . nbitk . . . nbitN and N non-complementary input terminals, bit0, bit1. . . bitk . . . bitN. The nbit and bit terminals with corresponding numbers of bit write logic network 14 and bit write logic network 16 are connected to identical vertically extending lines (i.e., leads) in SRAM 10 with line numbers corresponding to the terminal numbers so that, e.g., the nbitk terminal of write logic network 14 and the nbitk terminal of bit read logic network 16 are connected to the nbitk line of SRAM 10, while bitk terminal of write logic network 14 and bitk terminal of read logic network 16 are connected to the bitk line of SRAM 10.
FIG. 2 is a circuit diagram of the structure included in each cell of SRAM 10. For purposes of illustration, FIG. 2 is considered to be the cell at wordi bitk. The cell of FIG. 2 includes two inverters 20 and 22, connected to each other in a regenerative, positive feedback circuit 24. Inverter 20 includes complementary P-channel transistor 26 and N-channel transistor 28 having source drain paths connected in series between the chip positive DC power supply voltage (+VDD) and ground so that common drain terminals of FETs 26 and 28 are connected to terminal 30. Similarly, inverter 22 includes P-channel FET 32 and N-channel FET 34 having series connected source drain paths connected together between +VDD and ground so that FETs 32 and 34 have a common drain connection at terminal 36. The common drain terminal 30 of FETs 26 and 28 drives the gate electrodes of FETs 32 and 34 in parallel while the common drain terminal 36 of FETs 32 and 34 drives the gate electrodes of FETs 26 and 28 in parallel. Terminal 30 is selectively connected through the source drain path of N-channel pass gate FET 40 to nbitk line 42, while the source drain path of N-channel pass gate FET 44 selectively connects terminal 36 to bitk line 46. Wordi line 48 drives the gate electrodes of FETs 40 and 44 in parallel.
The cell of FIG. 2 was originally designed for differential read, differential write operation but has more recently been proposed for single-ended read, differential write operation. The widths and lengths of the dielectric regions of the gates of N-channel FETs 28 and 34 are the same, while the widths and lengths of the dielectric regions of the gates of P-channel FETs 26 and 32 are the same. Consequently, the gate widths and drive currents of FETs 26 and 36 are the same, while the gate widths and drive currents of FETs 28 and 34 are the same. In addition, the gate widths and gate capacitances of FETs 40 and 44 are the same because the lengths and widths of the dielectric regions of FETs 40 and 44 are the same. In a prior art configuration, the gate widths of N-channel FETs 28 and 34 are 0.36 xcexcm, the gate widths of P-channel FETs 26 and 32 are 0.32 xcexcm, and the gate widths of FETs 40 and 44 are 0.28 xcexcm and all of FETs 26, 28, 32, 34, 40 and 44 have gate lengths of 0.16 xcexcm.
When the cell of FIG. 2 is written or read for single-ended operation, logic network 12 supplies a positive (+VDD) voltage to line 48, while logic network 14 supplies signals to lines 42 and 46, and read logic network 16 supplies enable signals to its input terminals connected to lines 42 and 46 as follows. Prior to logic network 12 supplying the positive voltage to word line 48, write logic network 14 precharges one of nbit line 42 or bit line 46 to the positive voltage. To write a binary one into the cell of FIG. 2 logic network 14 reduces the voltage on nbit line 42 to ground but precharges the voltage on bit line 46 high prior to logic network 12 applying a high voltage to word line 48. FETs 40 and 44 are enabled by the positive voltage on line 48 to pass the low and high voltages on lines 42 and 46 to terminals 30 and 36. The low and high voltages at terminals 30 and 36 cause turn on of FETs 28 and 32 and turn off of FETs 26 and 34. Because of the regenerative connections of inverters 20 and 22, FETs 28 and 32 stay on and FETs 26 and 34 stay off and the voltages at terminals 30 and 36 respectively remain low and high after logic network 12 reduces the voltage on word line 48 to turn off FETs 40 and 44. Similarly, but in an opposite manner, logic network 14 writes a binary zero into the cell of FIG. 2 by applying high and low precharge voltages to lines 42 and 46, causing turn on of FETs 26 and 34 and turn off of FETs 28 and 32 in response to a positive voltage on write line 48. Hence, terminals 30 and 36 are at high and low voltages as a result of the positive voltage on line 48.
To read the binary value the cell of FIG. 2 stores, read logic network 16 samples the voltage at terminal 36 by enabling an input terminal of a detection circuit (not shown) of the read logic network after line 46 has been precharged to a high (i.e., positive) voltage and while word write/read logic network 12 applies a high voltage to line 48. If the cell of FIG. 2 is storing a bit causing terminals 30 and 36 to be respectively at high and low voltages, the low voltage at terminal 36 pulls down the voltage on line 46 when the positive voltage on line 48 turns on FET 44. The high voltage at terminal 36 causes line 46 to stay at a high voltage when the positive voltage on line 48 turns on FET 44. The detection circuit connected to line 46 senses the low and high voltages on line 46 to indicate the binary value the cell of FIG. 2 stores.
Because the memory of FIG. 1 and the cell of FIG. 2 are read on a single ended basis, the voltage variations of pulses on nbit line 42 during a read operation are not detected by bit read logic network 16. Consequently, a slow slew rate on nbit line 46 does not have a particularly adverse effect on read operations of the memory of FIG. 1 and the cell of FIG. 2. (Slew rate is the time rate of change of the voltage of leading edges of the pulses.)
FIG. 3 is a top view of the actual physical layout of the cell schematically illustrated in FIG. 2. The cell of FIG. 3 is approximately rectangular, having a width of 2.22 xcexcm, and a length of 2.52 xcexcm., i.e., the distance between parallel sides 50 and 52 is 2.22 xcexcm, while the distance between parallel sides 54 and 56, which extend at right angles to sides 50 and 52, is 2.52 xcexcm.
It is an object of the present invention to provide a new and improved SRAM cell.
An added object is to provide a new and improved SRAM having cells which are read by single-ended techniques.
Another object of the invention is to provide a new and improved SRAM cell that is read by single-ended techniques and has a smaller area than prior art SRAM cells.
An additional object of the invention is to provide a new and improved single-ended read SRAM cell having a faster bit slew read rate than prior art single-ended read SRAM cells.
A further object of the invention is to provide a new and improved SRAM having reduced access delay so it is capable of higher frequency operation read by single-ended techniques having a high bit line slew rate.
We have realized the performance of the cell illustrated in FIG. 2, when operated as described for differential write and single-ended read, can be improved to have a high read slew rate by changing the gate widths (i.e., the width of an oxide layer between the FET gate electrode and the semiconductor body including the FET source and drain electrodes) of FETs 28 and 34 so they are not the same. We have found that the size of the cell can be reduced with such a change. In particular, we realized the performance of the cell illustrated in FIG. 2 can be improved by making the gate width of FET 34 greater than the gate width of FET 28. In coming to this conclusion, we realized that for differential write and single-ended read, the slew of pulses used to read the voltage stored at terminal 30 is not important but that the slew rate of pulses used to read the voltage stored at terminal 36 is important. This is because no read operation is performed for the voltage stored at terminal 30 in the single-ended read mode. Increasing the gate width of FET 34 and decreasing the width of FET 28 causes the drive current of FET 34 to be greater than the drive current of FET 28 as well as greater than that of the prior art symmetric cell, i.e., greater current flows from the source to the drain of FET 34 than flows from the source to the drain of FET 28. A lower impedance is thereby established between bit line 46 and ground through the source drain path of FET 44 and the source drain path of FET 34 than between nbit line 42 and ground through the source drain paths of FETs 40 and 28. The faster slew rate of bit line 46 means higher frequency operation, i.e., lower SRAM access delay, than during a read operation with the prior art cell. The lower SRAM access delay enables a greater number of rows in SRAM matrix 10 to be accessed during a read operation in the same time period as the prior art, to thereby provide greater cell density.
The above and still further objects, features and advantages of the present invention will become apparent upon consideration of the following detailed description of a specific embodiment thereof, especially when taken in conjunction with the accompanying drawings.